Journal of Vibration Engineering & Technologies

https://doi.org/10.1007/s42417-022-00518-8

ORIGINAL PAPER

Coexistence of Attractors and Its Control with Selection

of a Desired Attractor in a Model of Extended Hindmarsh–Rose
Neuron with Nonlinear Smooth Fitting Function: Microcontroller
Implementation
Jules Tagne Fossi1,6 · Hélène Carole Edima2 · Zeric Tabekoueng Njitacke3,7 · Florent Feudjio Kemwoue1,6 ·
Joseph Marie Mendimi4 · Jacques Atangana5

Received: 2 October 2021 / Revised: 5 April 2022 / Accepted: 8 April 2022

© Krishtel eMaging Solutions Private Limited 2022

Abstract
Purpose  In this paper, we present a fitted model of the extended Hindmarsh–Rose (eHR) neuron constructed using the
hyperbolic tangent composite functions to replace the non-linear polynomial functions present in the original model of the
eHR neuron.
Methods  Thus, through classical techniques of analysis of nonlinear systems, we observe complex phenomena generated
by this system when the excitation current varies. In addition, a variation of the bifurcation parameters allows us to see that
the adjusted model of the neuron eHR is sensitive to the initial conditions, and therefore exhibits the multistability, which is
produced by the phenomenon of parallel branches or hysteresis in this system. To control these coexistences of attractors, a
control method based on the feedback term is applied to the model.
Results  We see that the addition of this space-dependent feedback term to the dynamic equation of this model drives the
dynamic towards the desired attractor by annihilating the other. This powerful technique makes it possible to move from a
multistable system to a monostable system.
Conclusion  Finally, we propose an on-board system implementation of this neural circuit using microcontroller technology.
This constitutes an important and reliable tool which can best mimic biological neurons.

Keywords  Multistability · Extended Hindmarsh–Rose (eHR) neuron · Control of coexisting attractors · Microcontroller

Hélène Carole Edima, Zeric Tabekoueng Njitacke, Florent Feudjio

Kemwoue, Joseph Marie Mendimi and Jacques Atangana have
contributed equally to this work.

2
* Jules Tagne Fossi National School of Agro‑Industrial Sciences, Food
jules_fossi@yahoo.fr Microbiology and Biotechnology Laboratory, University
of Ngaoundéré, Ngaoundéré, Cameroon
* Jacques Atangana
3
atanganajaques@yahoo.fr Department of Electrical and Electronic Engineering, College
of Technology (COT), University of Buea, P.O.Box 63, Buea,
Hélène Carole Edima
Cameroon
edimacarole@yahoo.fr
4
Department of Morphological Sciences and Morbid
Zeric Tabekoueng Njitacke
Anatomy, Faculty of Medicine and Biomedical Sciences,
zerictabekoueng@yahoo.fr
Yaoundé, Cameroon
Florent Feudjio Kemwoue 5
Department of Physics, Higher Teacher Training College
kemwoueflorent@yahoo.com
Yaoundé, University of Yaoundé I, Yaoundé, Cameroon
Joseph Marie Mendimi 6
Centre d’Excellence Africain des Technologies de
mendimajo@gmail.com
l’Information et de la Communication (CETIC) Université de
1 Yaoundé I, Yaoundé, Cameroon
Laboratory of Energy‑Electric and Electronic Systems,
7
Department of Physics, Faculty of Science, University Department of Automation,Biomechanics and Mechatronics,
of Yaoundé I, P.O. Box 812, Yaoundé, Cameroon Lodz University of Technology, Lodz, Poland

13
Vol.:(0123456789)
 Journal of Vibration Engineering & Technologies

Introduction [47, 49, 50, 65] have proposed analog simulations on the
Pspice software of analog neuromorphic circuits such as
A neuron is a cell in the nervous system that specializes in Hopfield neural networks and HR neural networks cou-
communicating and processing information [2]. Its sensi- pled by an asymmetric electrical synapse [51, 52]. These
tivity to various external stimuli allows it to fully play its neuron models, due to their complexities to non-linearity
role to respond appropriately and quickly [6, 38, 44]. As a (Multipliers), are generally very difficult to implement using
result, several mathematical models of neurons have been analog components. So, it is better to use other devices
presented to study the dynamic behaviors of the electrical such as microcontrollers [4] and FPGAs [14, 25, 26, 60,
activity of neurons, more precisely the mode of selection of 64]. Thus, authors such as Bao et al. [4], showed that in
neurons. The Hodgkin–Huxley neuron [19, 54, 69], the two- the Morris–Lecar autonomous neuron, the dynamics exhibit
and three-dimensional Hindmarsh–Rose (HR) neuron [17, chaotic bursting and coexisting multistable firing patterns.
18], the Lecar Morris neuron [21, 45], the Izhikevich neuron They also carried out experimental studies using a micro-
[70], the Hopfield neural network [47, 49, 50], the photosen- controller to present the different phenomena existing in the
sitive FitzHugh–Nagumo neuron [71, 73], and many others model. In addition, Rahimian et al. [60] implemented a two-
are among these models. In addition, electronic models [23, compartment Pinsky–Rinzel pyramid neuron model using
36] have also been proposed with the aim of developing digital FPGAs. Continuing along the same lines, Heidarpur
brain-type equipment that can, thus, effectively help in the et al. [14] implemented a neuromorphic online learning plat-
development of applications in artificial neural networks form of peak timing dependent plasticity on FPGA using the
[22]. Behaviors such as resting, spiking, bursting and cha- Izhikevich neuron.
otic states are commonly observed in some of these models Inspired by the above work and more particularly by the
[13]. Theoretical and in-depth studies have been carried out work of Cai et al. [24] on the analog implementation with-
including bifurcation analysis, Multistability analysis, syn- out multiplier of the 2D/3D HR neuron model, Bocheng
chronization transition and selection of models on the neural Bao et al. [5] focusing on the implementation of a mem-
network [3, 30–33, 43, 47, 48, 50–53, 68, 71–73]. ristive neuron model with an adapting synapse designed
The first mathematical model introduced by Hodgkin and from a fitting activation function circuit and Megam et al.
Huxley (HH) [19] is a very complex model taking a lot of [43] on bifurcation and multistability in the neural oscilla-
time in numerical simulations. For this reason, simpler neu- tor eRH, We propose in this article to present a new model
ral models have been developed. For example, Selverston of eHR neuron without a multiplier due to the complexity
et al. [61] studied a computer and electronic model of neu- of implementation of the original eHR model with a mul-
rons of the stomatogastric ganglion (STG). From these stud- tiplier and the cost of these multipliers. In the work of Cai
ies, they observed that biological neurons could be modu- et al, the authors show numerically and experimentally that
lated with only three or four degrees of freedom [61]. By using nonlinear composite hyperbolic tangent functions as a
focusing on the three-dimensional HR model [18], present- replacement for polynomial functions in the original models
ing itself as the model that produces several modes of startle of 2D/3D HR neurons, the fitted models reproduce practi-
behavior observed in biological neurons, its parameteriza- cally the same behaviors as the original models. Thus, we
tion space for chaotic activity is much more limited than that make use of the hyperbolic tangent functions as defined in
observed in neurons. For this, the authors proposed a modi- [24] to fit the nonlinear polynomial functions present in the
fied version of this model by adding a fourth term (slower original eHR neural oscillator. Hence, the eHR neuron fit
process) representing the dynamics of calcium [61]. Thus, model can be easily implemented using analog or digital
the Extended Hindmarsh–Rose (eHR) Neural Oscillator is components. Thus, to shed more light on the dynamic com-
a simple mathematical model, and is able to reproduce the plexity of this neural system, our objectives in this work are
complex dynamic behavior (spiking, bursting and chaotic) defined as follows:
of the central pyloric generators of the lobster stomatogastric
system [7, 61]. i. Define a novel adjusted eHR neuron model that is free
In addition, the manufacture of electronic devices that of cubic and quadratic nonlinearity (multipliers).
best mimic biological neurons and coupled neural networks ii. Identify the region in the parameter space in which the
is attracting much attention from researchers. Thus, many model develops the coexistence of multiple attractors.
analog and digital neuromorphic circuits have been realized iii. Control the coexisting attractors using the method
using analog components and digital devices [4, 14, 26, 34, of the feedback term that enables the selection of the
35]. Bao et al. [4], used the different analog components desired attractor.
to design and optimize an analog circuit to demonstrate an iv. Propose a microcontroller implementation of the pro-
adaptive neuron model based on synapses. Njitacke et al. posed model to validate our results.

13
Journal of Vibration Engineering & Technologies

This scientific contribution is organized as follows: in ⎧ ẋ = −F(x) + y + I − z

Sect. 2, we present the eHR neural oscillator fitted by two ⎪ ẏ = −G(x)
⎨ � � − y �− ew� (2)
nonlinear adjustment functions and perform the linear stabil- ż = 𝜇 �s �x − x0 �− z �
⎪
ity analysis. In Sect. 3, traditional nonlinear diagnostic tools ⎩ ẇ = 𝜈 f y + y0 − pw ,
such as bifurcation diagrams, Lyapunov spectrum graphs
and phase portraits are exploited to highlight complex phe- where the functions F(x) and G(x) represent cubic and quad-
nomena such as the coexistence of multiple attractors. Then, ratic polynomials, respectively.
we present a control method based on the feedback term to The production of cubic and quadratic polynomials gener-
control these attractor coexistence phenomena in Sect. 4. ally requires multipliers. Considering the very high cost of
Section 5 presents the experimental results obtained from this component, the analog circuit with multiplier becomes
the microcontroller implementation. While the conclusion very expensive than that without multiplier. For this reason,
is given in Sect. 6. it is important to propose a new model of the eHR neuron
that does not take into account the multipliers. Drawing on
the work of Cai et al. [24], we use the hyperbolic tangent
Description and Dynamic Analysis (Tanh(⋅)) mathematical function to fit the two nonlinear poly-
of the eHR Neuron Model nomials F(x) and G(x) of the eHR neuron model.
The cubic (F(x)) and quadratic (G(x)) polynomials are
Mathematical Expression of the Fitted Model replaced in the eHR model, respectively, by the functions
of the Neuron eHR ( H1 (x) ) and ( H2 (x) ) all defined from the composite hyper-
bolic tangent functions and described as follows.
The reduction of the model physiologically proposed by
Hodgkin and Huxley [19], made it possible to demonstrate a ⎧ H (x) =𝛼 tanh �X � + 𝛼 tanh �X � − 𝛼
⎪ 1 1 � �1 2 2 3
nonlinear mathematical model of the eHR neuron [61], used ⎨ tanh X�3 −� 𝛿01 � � (3)
for numerical simulations. This model presents many behav- ⎪ H2 (x) =𝛼4 tanh X4 − 𝛼5 tanh X5 + 𝛿02 ,
⎩
iors observed in biological neurons [15] and is described as
follows: with Xi = x𝜅i − 𝛿i with i = 1, 2, 3, 4, 5.
3 2 Here, the control parameters 𝛼i , 𝜅i and 𝛿i are used, respec-
⎧ ẋ = y − ax + bx + I − z tively, to determine the amplitude, gradient and horizontal
⎪ ẏ = c − dx − y − ew
2
⎨ �� � � (1) offset of the hyperbolic tangent function. 𝛿01 and 𝛿02 also
⎪ ż = 𝜇� �s x − x�0 − z �
denote the control parameters used to adjust the functions
⎩ ẇ = 𝜈 f y + y0 − pw .
H1 (x) and H2 (x).
Here, a, b, c, d, e, f , s, p, 𝜇, 𝜈, x0 and y0 , are parameters that By replacing, respectively, in system (2) the functions
govern the dynamics of the neuron. The parameter I rep- (F(x)) and (G(x)) by the functions ( H1 (x)) and ( H2 (x)) and
resents the excitation current of the neuron. The state vari- then considering the parameters defined in the previous sec-
ables x, y, z and w, respectively, represent the membrane tion, we can deduce a fit model of the eHR neuron defined
potential, the recovery variable associated with the fast cur- by:
rent, the slowly changing adaptation current and the even � � � �
⎧
slower dynamic process compared to the variable z for ⎪ ẋ = − 𝛼1 �tanh� X1 − 𝛼2 tanh X2 + 𝛼3
𝜈 < 𝜇 . It is very important to point out that the parameters ⎪ tanh X3 + 𝛿01 + y − z + I
𝜇 and 𝜈 are of paramount importance in neuronal activity. ⎪ � � � �
⎪ ẏ = − ew − 𝛼4 tanh X4 + 𝛼5 tanh X5 − 𝛿02
𝜇 represents the ratio of the time scales between the fast ⎨ (4)
and slow fluxes through the membrane of the neuron and ⎪ −y
�� � �
𝜈 controls the rate of change of the slower dynamic pro- ⎪ ż =𝜇 s x − x − z
⎪ � � 0
� �
cess of the state variable w, in particular the exchange of ⎪ ẇ =𝜈 f y + y0 − pw .
calcium between the intracellular warehouse and cyto- ⎩
plasm [43, 63]. Regularly, for the eHR neuron, we consider: Here, the parameters 𝜇, 𝜈 and I are chosen as
a = 1, b = 3, c = 1, d = 5, e = 0.03, f = 3, p = 0.96, s = 4, x0 = −1.6 and
control parameters of the eHR neuron fitting
y0 = 1.6. model. The other parameters are set as follows:
By setting F(x) = ax3 − bx2 and G(x) = dx2 − c the sys- 𝛼1 = 38.7, 𝜅1 = 0.7, 𝛿1 = −1.8, 𝛼2 = 38.7, 𝜅2 = 0.7,
tem (1) becomes: 𝛿2 = 3.2, 𝛼3 = 6, 𝜅3 = 0.8, 𝛿3 = 0.8, 𝛿01 = 2, 𝛼4 = 18,
𝜅4 = 0.98, 𝛿4 = 1.74, 𝛼5 = 18, 𝜅5 = 0.98, 𝛿5 = −1.74, 𝛿02 = 32.9.

13
 Journal of Vibration Engineering & Technologies

Equilibrium Point and Stability Analysis According to the Routh–Hurwitz criterion, the real parts of
the roots of Eq. (7) are positive if and only if:
To obtain the equilibrium points of the fitted model of the
m1 > 0,
eHR neuron described by the system Eq. (4), we revive this
system by setting ẋ = ẏ = ż = ẇ = 0 . Thus, we notice that m1 m2 − m3 > 0,
(9)
the system is a transcendental of the fourth order and can be m1 (m1 m3 − m1 m4 ) − m23 > 0,
solved numerically.
( The Jacobian ) matrix around the equilib- m4 > 0.
rium point Se = xe , ye , ze , we is described by:
Let us consider subsequently that the values of the param-
⎡ J11 1 −1 0 ⎤ eters 𝜇 and 𝜈 are fixed, respectively, at 0.009 and 0.007.
⎢J −1 0 −e ⎥
JSe = ⎢ 21 (5) Depending on the different values of the parameter I, the
0 − 𝜇 0 ⎥⎥
,
⎢𝜇s stability of each point of equilibrium is obtained by solving
⎣ 0 𝜈f 0 −𝜈p⎦ the characteristic polynomial Eq. (7). The points of equilib-
where rium as well as the eigenvalues which characterize the stabil-
ity of these points are calculated using MATLAB numerical
( ) ( )
J11 = − 𝛼1 𝜅1 sech2 X̄ 1 − 𝛼2 𝜅2 sech2 X̄ 2 simulators and grouped together in Table 1.
( ) It should be noted that Eq. (5) has only one point of equi-
+ 𝛼3 𝜅3 sech2 X̄ 3 , (6)
( )
2 ̄
( )
2 ̄
librium when the excitation current I is varied Fig. 1. There-
J21 = − 𝛼4 𝜅4 sech X4 + 𝛼5 𝜅5 sech X5 , fore, when I < 4.02 , we observe an unstable saddle point
( ) ( ) with two complex conjugate roots with positive real parts
with sech2 X̄ i = 1 − tanh2 X̄ i and X̄ i = xe 𝜅i − 𝛿i for
and two negative real roots. For I > 4.02 , we have a stable
i = 1, 2, 3, 4, 5
The coefficients J11 and J21 defined in Eq. (6) denote the
elements of the first column, first row and second row of the
Jacobian matrix, respectively.
The resolution of the characteristic polynomial of equa-
tion P(𝜆) = det(JSe − 𝜆Id ) , makes it possible to obtain four
eigenvalues with Id an identity matrix.

P(𝜆) = 𝜆4 + m1 𝜆3 + m2 𝜆2 + m3 𝜆 + m4 , (7)

where
m1 = 𝜈 p − J11 + 𝜇 + 1,
( ) ( )
m2 = 𝜈 p − J11 + s + 1 𝜇 + −J11 p + ef + p 𝜈
− J11 − J21 ,
((( ) ) ) (8)
m3 = s − J11 + 1 p + ef 𝜈 + s − J11 − J21 𝜇
(( ) )
− 𝜈 J11 + J21 p + J11 ef ,
(( ) ( ))
m4 = 𝜈 s − J11 − J21 p + ef s − J11 𝜇.
Fig. 1  Variation of the equilibrium point under different values of the
excitation current I. The blue segment indicates the unstable zone and
the red one the stable zone

Table 1  DC equilibrium points, eigenvalues and stability

I Equilibrium points Ē Eigenvalues 𝜆1 , 𝜆2,𝜆3 and 𝜆4 Stability

0.5 {−1.2843, −6.4806, 1.2624, −15.2520} 𝜆1,2 = 0.0384 ± 0.0317j, 𝜆3 = −0.0076, 𝜆4 = −13.8146 Unstable
1.5 {−1.0243, −3.5797, 2.3027, −0.7970} 𝜆1,2 = 0.0458 ± 0.0327j, 𝜆3 = −0.0076, 𝜆4 = −10.4439 Unstable
2.5 {−0.7685, −1.5416, 3.3257, −0.1823} 𝜆1,2 = 0.0327 ± 0.0598j, 𝜆3 = −0.0075, 𝜆4 = −7.5667 Unstable
3 {−0.6456, −0.8268, 3.8175, 2.4161} 𝜆1,2 = 0.0228 ± 0.0720j, 𝜆3 = −0.0075, 𝜆4 = −6.3412 Unstable
3.5 {−0.5250, −0.2738, 4.2996, 4.1441} 𝜆1,2 = 0.0120 ± 0.0832j, 𝜆3 = −0.0074, 𝜆4 = −5.2294 Unstable
4 {−0.4060, −0.1423, 4.7759, 5.4446} 𝜆1,2 = 0.0002 ± 0.0943j, 𝜆3 = −0.0074, 𝜆4 = −4.2110 Unstable
4.5 {−0.2876, 0.4396, 5.2493, 6.3738} 𝜆1,2 = −0.0142 ± 0.0943j, 𝜆3 = −0.0074, 𝜆4 = −3.2628 Stable

13
Journal of Vibration Engineering & Technologies

saddle point with two complex conjugate roots with negative

real parts and two negative real roots.
Looking at Fig. 1, we see that the stability of the fitted
model is a function of the excitation current. By observing
the characteristic polynomial equation P(𝜆) we find that the
stability of the fitted model around the equilibrium point
depends mainly on the ratio of the time scales between the
fast and slow flows (𝜇) and on the controller of the rate of
variation of the dynamic process of the variable w (𝜈) . There-
fore, it is important to study the effect of these parameters on
the stability of the system, which will allow us to highlight
and explain the various local bifurcation phenomena.
Consider that the parameters 𝜇 and 𝜈 vary, respectively,
from 0 < 𝜇 < 0.04 and 0 < 𝜈 < 0.15 . Figure 2 shows us that
these variables influence the stability of the system around
the equilibrium point. The blue zone in this figure represents
pairs of coordinates (𝜈, 𝜇) for which a stable or quasi-stable Fig. 3  Representation of the eigenvalues, solution of the charac-
equilibrium point is observed, while the red zone represents teristic polynomial Eq.  7, in the complex plane ( Re(𝜆) , Im(𝜆) ) for
pairs of coordinates for which an unstable equilibrium point 0 < 𝜈 < 0.15 , 𝜇 = 0.009 and I = 3 . Provided that JSe is a real matrix,
complex eigenvalues occur in complex conjugate pairs responsible of
is observed. In view of these equilibrium point stability tran-
the symmetry observed along the real axis. The locus intersects the
sitions, we can affirm that the solutions of Eq. (7) undergo imaginary axis and, thus, suggests the possibility of Hopf bifurcation
qualitative changes.
For the rest, we fix 𝜇 = 0.009 . Figure 3 gives a repre-
sentation of the solutions of the eigenvalues in the complex Bifurcation and Coexistence of Multiple
plane for the values of 𝜈 varying from 0 to 0.15. The sym- Attractors
metrical presence of complex pairs of conjugate eigenvalues
along the real axis shows a possibility of the existence of a To highlight the different complex phenomena that can act
Hopf bifurcation [28, 67]. in the fitted model of the eHR neuron, we solve numerically
using Turbo Pascal the nonlinear system 4 using the fourth-
order Runge–Kutta algorithm with a time step h = 0.001 for
a good precision.

Bifurcation and Road to Chaos

Generally, the quantitative and qualitative analysis of the

bifurcation diagram, the Lyapunov spectra, the portraits of
phases and the frequency spectra make it possible to char-
acterize and predict the evolution of dynamic behavior. The
bifurcation diagram and the Lyapunov spectra are calculated
as a function of a single parameter called control or bifur-
cation. We focused on the bifurcation parameters that can
be manipulated experimentally, for example, the excitation
current I injected into the system (5) and the rate of change
of the slower dynamic process 𝜈 . First, we consider the val-
ues of the parameters of the system defined in Sect. 2 with
𝜇 = 0.009 and 𝜈 = 0.007 . Through Fig. 4, we present the
bifurcation diagram with respect to I well as its Lyapunov
exponent.
It is important to mention that the bifurcation diagrams
are obtained by considering the maxima of the membrane
Fig. 2  Basin of stability of the equilibrium point Se , in the plane
( 𝜈 , 𝜇 ) showing the region where this equilibrium point is stable potential x(t) with a sufficiently long transient time. Note
or quasi-stable (blue) and the region where it is unstable (red) for also that: if 𝜆max < 0 , the system evolves towards a regular
0 < 𝜇 < 0.04 and 0 < 𝜈 < 0.15 , with I = 3 state, if 𝜆max = 0 , it evolves towards a torus state and when

13
 Journal of Vibration Engineering & Technologies

2 𝜆max > 0 , the chaotic state is observed. Therefore, when we

look at these numbers, we can identify different transitions
1 to the chaotic state such as period doubling or crisis scenar-
ios [43]. For I = 3.5 , the corresponding maximal Lyapunov
x

0 exponent is 𝜆max = 0.0004 . Seeing that it is positive, we can

conclude that there is a chaotic behavior in this model and
-1 many other very rich dynamic phenomena such as a continu-
0 1 2 3 4
I ous crisis and a dynamic structured in blocks as it was dis-
10-3
covered in the work of Megam et al. [43] for the eHR model
and Gonzales-Miranda [8, 9] for the original HR model.
10 This internal crisis characterizing the transition between the
Spiking regime and the collapsing regime [1, 10–12, 46] is
max

5 observed for a value of I ≈ 2.36 . Moreover, we observe the

bifurcation structure of a cascade of doubloons of reversed
0 period as shown in Fig. 4.
0 1 2 3 4 In the time series in Fig. 5, we show that proper set-
I tings of the bifurcation parameter, allow us to have excit-
ing behaviors such as burst state Fig. 5a and chaotic flows
Fig. 4  Bifurcation diagram (a) showing the local maxima of the Fig. 5b. Figure 5c, d, respectively, show the phase portraits
membrane potential × as a function of the external excitation cur- in the plane (z − x) of the time series of Fig. 5a, b. In view
rent I and in (b) the corresponding graph of the maximal Lyapu- of these results, we can say, that this model can function in
nov spectrum. These graphs are obtained with the initial condition
(0.1,  0,  0,  0). The other values of the parameters are defined in the
an efficient way like the model of the eHR neuron.
text

2 2

1.5 1.5
1
1
0.5
0.5
0
x

0
-0.5
-0.5
-1

-1.5 -1

-2 -1.5
9.5 9.52 9.54 9.56 9.58 9.6 9.62 9.64 9.66 9.68 9.7 9.5 9.52 9.54 9.56 9.58 9.6 9.62 9.64 9.66 9.68 9.7
t 104 t 104

(a) (b )

(c) (d )

Fig. 5  Temporal evolution of the membrane potential x(t) and the for- (b, d) show a chaotic bursting for I = 3.5 . They are obtained with the
mation of its attractor in the plane (z − x) of the fitted model of the same condition and the same values of the parameters as defined in
eHR neuron for different values of the applied excitation current I. Fig. 4
The graphs (a, c) show a periodic bursting for I = 3 , while the graphs

13
Journal of Vibration Engineering & Technologies

Coexistence of Multiple Attractors allows to obtain the periodic attractor and (0.1, 0, 0, 0) for
the chaotic attractor. This typical type of periodic and cha-
The coexistence of several attractors also called multistabil- otic attractor coexistence is very important in the study of
ity is a phenomenon widely studied in nonlinear dynamic brain dynamics because it can be used to justify an epileptic
systems such as electronic circuits [27, 39, 59, 66], lasers seizure that coexists with a normal state of the brain. In this
[42] and biological systems [40, 41, 56, 57, 62]. The aim particular case, the epileptic seizure corresponds to peri-
of this subsection is to study the phenomenon of multista- odic behavior, while the normal state corresponds to chaotic
bility in the adjusted eHR neural system given in Sect. 2.2 behavior [47, 55].
(Eq. (5)). For this, several bifurcation diagrams are calcu- Second, we propose to demonstrate another phenomenon
lated using different techniques related to parallel branches of coexistence of solutions with two competing attractors
and hysteresis. First, we vary the excitation current I in the using hysteretic dynamics. Thus, we vary the control param-
range 3.58 < I < 3.63 and obtain the bifurcation diagrams eter of the rate of change of the slower dynamic process of
of Fig. 6, which show the coexistence of several attractors the state variable w denoted by 𝜈 . In Fig. 8a, we present the
due to parallel branches. bifurcation diagram and the corresponding Lyapunov expo-
In this figure, we have represented the local maxima of nent in Fig. 8b when 𝜈 is taken in the interval 0 < 𝜈 < 0.13 .
the potential of the membrane × with respect to I which Figure 8c, d show the enlargement of the bifurcation dia-
is the parameter linked to the excitation current of the gram as well as the Lyapunov exponent of Fig. 8a, b. The
neuronal system. Diagram (a) is obtained by decreasing I diagram in blue is obtained by increasing the parameter 𝜈
in the interval 3.58 < I < 3.63 with a fixed initial condi- and the one in red by decreasing the same parameter.
tion (0.1, 0, 0, 0). In (b), the diagrams in blue and red are In view of these graphs, we note that two attractors can
obtained by reducing the parameter I under different initial coexist in the proposed model. Considering the value of the
conditions. For these diagrams, the final state of each itera- parameter 𝜈 = 0.1015 taken on the bifurcation diagram, we
tion is used as the new initial condition for the next iteration. show through Fig. 9a that two periodic attractors can coexist:
The other parameters are defined as in Fig. 4. From Fig. 6b, the attractor of period-4 is captured with the initial condition
one can observe several windows of divergence between the (0.2, 0, 0, 0) and the attractor of period-12 is captured with
superimposed diagrams thus justifying the phenomenon of the initial condition (1.1, 0, 0, 0). In neuroscience, this case
coexistence of attractors coming from the bifurcation of par- corresponds in cerebral dynamics to the coexistence between
allel branches. For discrete values of the parameter I taken two pathological states because, in cerebral dynamics, the
on the bifurcation diagram of Fig. 6b, we show through normal state is chaotic [55]. Moreover, when 𝜈 = 0.102 ,
Fig. 7 the coexistence of two attractors. For the values of we observe as shown in Fig. 9b the coexistence of two dif-
I = 3.592, 3.594, 3.612 and 3.615, we obtain, respectively, an ferent asymmetric attractors among which we have a peri-
attractor of period-12 (Fig. 7a), period-8 (Fig. 7b), period-4 odic attractor and a chaotic attractor. The periodic attractor
(Fig. 7c), and period-2 (Fig. 7d), each of which coexists admitting a 𝜆max = −4.09 × 10−4 is obtained with the initial
with a chaotic attractor. The initial condition (0.1, 0, 0.1, 0) condition (0.8, 0, 0, 0) while the chaotic attractor admitting

1 1.5

0.8
1
0.6

0.4 0.5
x

0.2
0
0

-0.2 -0.5
3.58 3.59 3.6 3.61 3.62 3.63 3.58 3.59 3.6 3.61 3.62 3.63
I I
(a) (b)

Fig. 6  Bifurcation diagram (a) showing local maxima of mem- ditions. The diagram in blue is obtained with the initial condition
brane potentials x(t) as a function of parameter I in the range (0.1, 0, 0.1, 0), while the one in red is obtained for the initial condi-
3.58 < I < 3.63  , obtained with the initial condition fixed: tion (0.1, 0, 0, 0). They are obtained with parameters set as follows:
(0.1,  0,  0,  0). In (b), the blue and red diagrams are obtained by 𝜈 = 0.007 and 𝜇 = 0.009
decreasing the parameter in the same range from different initial con-

13
 Journal of Vibration Engineering & Technologies

Fig. 7  Coexistence of two different asymmetric attractors (periodic red attractor, respectively: a I = 3.592 ; b I = 3.594 ; c I = 3.612 and
and chaotic) of different shape for discrete I values with initial con- d I = 3.615 . The same parameters of Fig. 6 are considered
dition (0.1,  0,  0.1,  0) for the blue attractor and (0.1,  0,  0,  0) for the

2
1
1
0.5
x

0 0

-1
0 0.02 0.04 0.06 0.08 0.1 0.12 0.101 0.1015 0.102 0.1025

10-3 10-3
4
10
2
max

max

0
0
0 0.02 0.04 0.06 0.08 0.1 0.12 0.101 0.1015 0.102 0.1025

Fig. 8  Bifurcation diagram (a) showing the local maxima of the b). The diagram in blue is obtained by increasing the value of 𝜈 and
membrane potential x(t) as a function of the parameter 𝜈 in the range the one in red is obtained by decreasing this value with the same ini-
0 < 𝜈 < 0.13 and in (b) the corresponding graph of the maximal Lya- tial condition (0.8, 0, 0, 0). They are obtained with parameters set as
punov spectrum. Graphs (c, d), respectively, represent the bifurcation follows: 𝜇 = 0.009 and I = 3
diagram and the maximum Lyapunov spectrum to zoom in from (a,

13
Journal of Vibration Engineering & Technologies

Fig. 9  Coexistence of two different asymmetric attractors of different 0) for the red attractor and c, d 𝜈 = 0.102 (a limit cycles of period-4
shape for discrete 𝜈 values as well as their corresponding time series: and chaotic) with initial condition (0.8, 0.0, -1, 0.0) for the blue
a, b 𝜈 = 0.1015 , (two limit cycles of period-4 and period-12) with ini- attractor and (0.8, 0, 0, 0, 0) for the red attractor. The same param-
tial condition (0.2 , 0.0, 0.0, 0.0) for the blue attractor and (1.1,0, 0, 0, eters of Fig. 8 are considered

{
a 𝜆max = 1.95 × 10−3 is obtained with the initial condition
�
1 if X ∈ R
(0.8, 0, 1, 0). We can conclude that the qualitative behavior
h(X) = � (11)
0 if X ∉ R ,
of long-term motion of a nonlinear dynamic system depends
where R ⊂ R is a subset of the complete state space and
′
greatly on the initial condition.
R is a subset of the system. The choice of depends on the
state space of the system considered and generally it will be
Control of Multistability of the Fitted Model chosen so that the trajectories of only one of the attractors
of the Neuron eHR (attractor to be eliminated) exist in R  . Thus, the feedback
′

term vanishes when the dynamics of the system is directed

The phenomenon of multistability in the study of brain from a given attractor to the desired attractor for the choice
dynamics is very important to justify an epileptic seizure of h (X) which allows the dynamic system to move from
(periodic behavior) which coexists with a normal state of the a multistable system to a system monostable for a set of
brain (chaotic behavior) [47, 55]. To eliminate the pathologi- selected parameters.
cal condition of the brain (epilepsy), it is important to con- Presently, let us apply this control method on the dynam-
trol this multistability [31]. Thus, on the basis of the work of ics of the fitted model of the eHR neuron. The main objec-
Kiran et al. [29], consider a multistable dynamic system with tive in this study is to remove the periodic attractor (patho-
m-dimensional, whose dynamics is defined by Ẋ = G(X) , logical state of the brain) to remain only with the chaotic
where X is an m-dimensional vector of state variables and attractor (normal state of the brain) without changing the
G(X) is the field vector describing its dynamics. Let us admit intrinsic dynamics of the system. Feedback coupling is intro-
that for a specific initial condition, the dynamics of the sys- duced along the variable × with the feedback strength as
tem presents a specific attractor. Thus, the control scheme shown in the Eq. (12).
for selecting an attractor has been proposed as follows
⎧ � � � �
Ẋ = G(X) − 𝜉h(X)C × (X − B), (10) ⎪ ẋ = − 𝛼1 �tanh� X1 − 𝛼2 tanh X2 + 𝛼3
⎪ tanh X3 + 𝛿01 + y − z + I − 𝜉h(x)(x − b)
{ } ⎪ � � � �
where 𝜉 is the feedback strength, C = Cij is an m × m ⎪ ẏ = − ew − 𝛼4 tanh X4 + 𝛼5 tanh X5 − 𝛿02
matrix, h(X) is a space dependent term, and the vector B ⎨ (12)
⎪ −y
represents the unstable fixed point in the neighborhood of �� � �
⎪ ż =𝜇 s x − x − z
the desired attractor.
{ To keep these dynamic properties, we ⎪ � � 0
� �
1 if i = j = 1 ⎪ ẇ =𝜈 f y + y0 − pw .
consider Cij = and Bi=1 = b . The function ⎩
0 otherwise
h(X) can be defined as

13
 Journal of Vibration Engineering & Technologies

1.5 1.5

1
1

0.5
0.5
0

0
-0.5

-1 -0.5
4 4.2 4.4 4.6 0 0.1 0.2 0.3 0.4 0.5

(a) (b)

(c)


Fig. 10  a The shaded region R of the periodic attractor represents the

′
the different initial conditions. The diagram in blue is obtained with
fraction of state space where the feedback term is active. b Bifurca- the initial condition (0.1, 0, 0, 0) while the one in red is obtained for
tion diagram showing the local maxima of the membrane potential x the initial condition (0.1, 0, 0.1, 0) with b = 2.7 . c State of the desired
as a function of the feedback strength 𝜉 in the range 0 ≤ 𝜉 ≤ 0.5 , from attractor (chaotic) for b = 2.7 and 𝜉 = 0.404 . The same parameters of
Fig. 8d are considered

In the absence of the controller, the system exhibits multista- the feedback strength 𝜉 > 0.404 , the dynamics of the fitted
bility phenomena as indicated in Sect. 3.2. Consider Fig. 7c model of the eHR neuron transits from a bistable system to
where we observe the coexistence of two attractors (periodic a monostable system.
and chaotic) with different initial conditions. We propose
to remove the periodic attractor, thus, we define a region of
the state space of the periodic attractor as shown in Fig. 10a. Hardware Implementation
The function h(x) is, therefore, defined by
{ For the sake of simplicity, speed, and precision, the imple-
1 if 1 < x < 1.2 mentation of this neural circuit has been designed around
h(x) =
0 Otherwise. (13)
an ESP32 microcontroller from ESPRESSIF. It is fully
compatible with the Arduino IDE (Integrated Development
Environment), WIFI, Bluetooth, DAC (Digital to Analog
For a value of b = 2.7 , it is shown through the bifurcation
Converter) and ADC (Analog to Digital Converter). It is
diagram of Fig. 10b that as the feedback strength increases,
equipped with a double processor, a 40 MHz crystal used as
the dynamics of the periodic attractor are substituted into
a crystal oscillator to synchronize the instructions of the pro-
chaotic attractor. Thus, monostability is achieved when
gram, 15 pins constitute a capacitive sensor and a thermom-
the value of the feedback strength is 𝜉 = 0.404 . Looking
eter and is supplied with a voltage of 5 V (see Fig. 11). He
at Fig. 10c, we assert that the periodic attractor switches
programming language used here in the Arduino software
to the chaotic attractor. We can conclude that for a value of

13
Journal of Vibration Engineering & Technologies

described by the ODE equations. Considering the param-

eters of Fig. 5, we observe through Fig. 13, the experimental
results of the burst state and the chaotic state of the system.
Figure 13a, b, give us the temporal evolution of the mem-
brane potential for the respective values of the excitation
current 3 and 3.5. Figures 13c, d show the various phase
portraits in the two-dimensional plane (z − x) associated
with Fig. 13a, b. In Fig. 14, we show experimentally that
the device can reproduce multistability phenomena. To
illustrate this phenomenon, we consider the case where we
have a chaotic attractor that coexists with a periodic attrac-
tor of period-2 as shown in Fig. 7d. We can see from these
experimental results a similarity with the numerical results.
Therefore, this microcontroller device is a reliable and rapid
tool for designing neuromorphic circuits capable of giving
an appropriate prediction of the dynamics of biological
Fig. 11  The ESP32-WROOM-32D microcontroller neurons.

IDE is “C / Arduino”, which is very closed and compatible Conclusion

with the “C” programming language.
To implement our neural model defined by the system In this paper, an eHR fitted neuron model was designed
(Eq. (4)) with the ESP32-WROOM-32D microcontroller, based on the composition of hyperbolic tangent nonlinear
we use the fourth-order Runge–Kutta numerical solution functions, replacing two nonlinear functions, one cubic
method with h = 0.001 a time step to discretize our contin- and the other quadratic, present in the original eHR neuron
uous-time equations.The presence of the DAC integrated in model [43]. This approach enables us to avoid the use of
the microcontroller allows us to recover at the output of the multipliers when implementing the analog circuit, which
pins analog quantities representing the state variables of the usually makes it very expensive. After this, we performed a
system. These analog quantities obtained are displayed on study of the bifurcations and multistability of the solutions
a computer using a fast data acquisition module designed of the eHR adjusted neural oscillator to show the existence
with an Arduino UNO board as can be seen in Fig. 12a, b. of complex phenomena in the system. Before addressing that
The above experimental setup is capable of reproducing work, a detailed stability analysis was applied to character-
the complex behaviors generated by biological neuron as ize the neural oscillator. Then, the excitation current has

Fig. 12  Block diagram (a) and

experimental setup (b) of the
microcontroller-based imple-
mentation of the eHR model

13
 Journal of Vibration Engineering & Technologies

Fig. 13  Experimental results of the temporal evolution of the mem- excitation current I: a, c I = 3 (periodic bursting with seven spikes
brane potential x(t) and attractor in the plane (z − x) of the fitted per burst) and b, d I = 3.5 (chaotic bursting with innumerable
model of the extended HR neuron for different values of the applied spikes). The other values of the parameters are defined in the text

Fig. 14  Experimental result of the coexistence of two asymmetric attractors of different shape for the value of I = 3.615 : a periodic attractor
with initial condition (0, 1, 0, 0.1, 0) and b chaotic attractor with the initial condition (0, 1, 0, 0, 0). The same parameters of Fig. 8 are considered

been varied, and it is found that the eHR neuron exhibits term dependent on the dynamic equation space has been
complex and varied behaviors when recording the membrane proposed. It has been found that when the feedback strength
potential. It was also observed that this model exhibits a reaches a certain threshold value, the model will drive the
sensitivity to the change of the initial conditions, which dynamics towards the desired attractor by annihilating the
indicates the existence of a multistability phenomenon. To other. Finally, an implementation of the adjusted neural
control this phenomenon, a method based on the feedback oscillator is implemented using a microcontroller. It is then

13
Journal of Vibration Engineering & Technologies

possible to obtain real electrical signals, thanks to a simple 15. Herz AVM, Gollisch T, Machens CK, Jaeger D (2006)Modeling
program inserted into a microcontroller. The experimen- single-neuron dynamics and computations: a balance of detail and
abstraction. Sci 314:80
tal results obtained are interesting and comparable to the 16. Heyward P, Ennis M, Keller A, Shipley MT (2001) Membrane
numerical results presented above, which use on-board tech- bistability in olfactory bulb mitral cells. J Neurosci 21:5311–20
nologies to generate the complex behaviors of the biological 17. Hindmarsh JL, Rose RM (1982) A model of the nerve impulse
neuron. It turns out that using the microcontroller appears to using two first-order differential equations. Nature 296:162–164
18. Hindmarsh JL, Rose RM (1984) A model of neuronal bursting
be a good way to display real electronic signals from neural using three coupled first order differential equations. Proc R Soc
circuits. Therefore, the results obtained provide potentially Lond B Biol Sci 221:87–102
useful information to further our knowledge of the function- 19. Hodgkin AL, Huxley AF (1952) Propagation of electrical signals
ing of the neural system and the encoding of information along giant nerve fibres. J Physiol Lond 117:500
20. Hounsgaard J, Kiehn O (1989) Serotonin-induced bistability of
[16, 20, 37, 58]. turtle motoneurones caused by a nifedipine-sensitive calcium pla-
teau potential. J Physiol 414:265–82
Acknowledgements  Jules Tagne Fossi thanks the Faculty of Sciences 21. Hu X, Liu C, Liu NJ, Li S (2016) An electronic implementation
of the University of Ngaoundéré for its important contribution. Zeric for Morris-Lecar neuron model. Nonlinear Dyn 84:2317–2332
Tabekoueng Njitacke has been supported by the Polish National Sci- 22. Hu X, Feng G, Duan SK, Liu L (2017) A memristive multilayer
ence Centre under the Grant OPUS 14 No. 2017/27/B/ST8/01330. cellular neural network with applications to image processing.
IEEE Trans Neural Netw Learn Syst 28:1889–1901
Declarations  23. Jiang YN, Huang P, Zhu DB, Han RZ, Liu LF, Liu XY, Kang
JF (2019) Design and hardware implementation of neuromorphic
systems with RRAM synapses and threshold controlled neu-
Competing interest  The authors declare that they have no known rons for pattern recognition. IEEE Trans Circ Syst I Reg Papers
competing financial interests or personal relationships that could have 65:2726–2738
appeared to influence the work reported in this paper. 24. Cai JM, Bao H, Xu Q, Hua ZY, Bao B (2021) Smooth nonlin-
ear fitting scheme for analog multiplierless implementation of
Hindmarsh-Rose neuron model. Nonlinear Dyn. https://​doi.​org/​
10.​1007/​s11071-​021-​06453-9
References 25. Jokar E, Soleimani H (2017) Digital multiplierless realization of
a calcium-based plasticity model. IEEE Trans Circ Syst II Exp
1. Abraham RH, Stewart HB (1986) A chaotic blue sky catastrophe Briefs 64:832–836
in forced relaxation oscillations. Phys D 21:394–400 26. Jokar E, Abolfathi H, Ahmadi A, Ahmadi M (2019) An efficient
2. Aram Z, Jafari S, Ma J, Sprott JC, Zendehrouh S, Pham VT (2017) uniform-segmented neuron model for large-scale neuromorphic
Using chaotic artificial neural networks to model memory in the circuit design: simulation and FPGA synthesis results. IEEE Trans
brain. Commun Nonlinear Sci Numer Simul 44:449–59 Circ Syst I Reg Papers 66:2336–2349
3. Bao BC, Yang QF, Zhu L, Bao H, Xu Q, Yu YJ, Chen M (2019) 27. Kengne J, Chedjou JC, Kom M, Kyamakya K, Kamdoum VT
Chaotic bursting dynamics and coexisting multistable firing pat- (2014) Regular oscillations, chaos and multistability in a system
terns in 3D autonomous Morris-Lecar model and microcontroller- of two coupled van der Pol oscillators numerical and experimental
based validations. Int. J. Bifurc Chaos 29:1950134 studies. Nonlinear Dyn 76:1119–32
4. Bao BC, Zhu Y, Li CQ, Bao H, Xu Q (2020) Global multistability 28. Kengne J, Jafari S, Njitacke ZT, Yousefi AK, Cheukem A (2017)
and analog circuit implementation of an adapting synapse-based Dynamic analysis and electronic circuit implementation of a novel
neuron model. Nonlinear Dyn 101:1105–1118 3D autonomous system without linear terms. Commun Nonlinear
5. Bao BC, Zhu YX, Ma J (2021) Memristive neuron model with an Sci Numer Simul. https://​doi.​org/​10.​1016/j.​cnsns.​2017.​04.​017
adapting synapse and its hardware experiments. Sci China Tech 29. Kiran Y, Neeraj KK, Manish DS (2017) Intermittent feedback
Sci 64(5):1107–1117. https://d​ oi.o​ rg/1​ 0.1​ 007/s​ 11431-0​ 20-1​ 730-0 induces attractor selection. Phys Rev E 95:042215
6. Borgers C (2017) An introduction to modeling neuronal dynamics. 30. Kuznetsov YA (1998) Elements of applied bifurcation theory.
Springer International Publishing, p 17 Springer, New YorK
7. De Lange E, Hasler M (2008) Predicting single spikes and spike 31. Lai Q, Norouzi B, Liu F (2018) Dynamic analysis, circuit reali-
patterns with the Hindmarsh-Rose model. Biol Cybern 99:349–60 zation, control design and image encryption application of an
8. Gonzalez-Miranda JM (2003) Observation of a continuous interior extended Lu system with coexisting attractors. Chaos Solitons
crisis in the Hindmarsh-Rose neuron model. Chaos 13:825–45 Fractals 114:230–245
9. Gonzalez-Miranda JM (2007) Complex bifurcation structures in 32. Lai Q, Wan Z, Kengne LK, Kuate PDK, Chen C (2021) Two-
the Hindmarsh-Rose model. Int J Bifurc Chaos 17:3071–83 memristor-based chaotic system with infinite coexisting attractors.
10. Grebogi C, Ott E, Yoke JA (1983) Crises, sudden changes in cha- IEEE Trans Circ Syst II Express Briefs 68(6):2197–2201
otic attractors and transient chaos. Phys D 7:181–200 33. Lai Q (2021) A unified chaotic system with various coexisting
11. Grebogi C, Ott E, Romeiras F, Yoke JA (1987) Critical exponents attractors. Int J Bifurc Chaos 31(1):2150013
for crisis induced intermittency. Phys Rev A 36:5365–80 34. Lai Q, Wan Z, Kengne LK, Kuate PDK (2020) Modeling and
12. Grebogi C, Ott E, Yorke JA (1987) Chaos, strange attractors and fractal circuit realisation of a new no-equilibrium chaotic system
basin boundaries in nonlinear dynamics. Sci New Ser 238:632–8 with hidden attractor and coexisting attractors. Electron Lett
13. Gu HG, Pan BB, Chen GR, Duan LX (2014) Biological experi- 56(20):1044–1046
mental demonstration of bifurcations from bursting to spiking 35. Lai Q, Wan Z, Kuate PDK, Fotsin H (2020) Coexisting attractors,
predicted by theoretical models. Nonlinear Dyn 78:391–407 circuit implementation and synchronization control of a new cha-
14. Heidarpur M, Ahmadi A, Ahmadi M, Azghadi MR (2019) otic system evolved from the simplest memristor chaotic circuit.
CORDIC- SNN: On-FPGA STDP learning with Izhikevich neu- Commun Nonlinear Sci Numer Simul 89:105341
rons. IEEE Trans Circ Syst I Reg Papers 66:2651–2661

13
 Journal of Vibration Engineering & Technologies

36. Larras B, Lahuec C, Seguin F, Arzel M (2016) Ultra-lowenergy 56. Pastor-Diaz I, Perez-Garcia VM, Encinas-Sanz F, Gerra JM
mixed-signal IC implementing encoded neural networks. IEEE (1983) Ordered and chaotic behavior of two coupled van der Pol
Trans Circ Syst I Reg Papers 63:1974–1985 oscillators. Phys Rev E 48:171–82
37. Loewenstein Y, Mahon S, Chadderton P, Kitamura K, Sompolin- 57. Pastor-Diaz I, Lopez-Fragas A (1995) Dynamics of two coupled
sky h (2005) Bistability of cerebellar Purkinje cells modulated by van der Pol oscillators. Phys Rev E 52:1480–9
sensory stimulation. Nat Neurosci 8:202–11 58. Perrier JF, Hounsgaard J (2000) Development and regulation of response
38. Ma J, Yang Z, Yang L et al (2019) Phase coupling synchronization properties in spinal cord motoneurons. Brain Res Bull 53:529–35
of FHN neurons connected by a Josephson junction. J Zhejiang 59. Prakash P, Rajagopal K, Singh JP, Roy BK (2018) Megastabil-
Univ Sci A 20:639 ity in a quasi-periodically forced system exhibiting multistabil-
39. Maggio GM, De Feo O, Kennedy MP (1999) Nonlinear analysis ity, quasi-periodic behaviour, and its analogue circuit simulation.
of the Colpitts oscillator and application to design. IEEE Trans AEU-Int J Electron Commun 92:111–115
Circ Syst 46:1118–30 60. Rahimian E, Zabihi S, Amiri M, Linares-Barranco B (2018)
40. Malashchenko T, Shilnikov A, Cymbalyuk G (2011) Bistability of Digital implementation of the two-compartmental PinskyRinzel
bursting and silence regimes in a model of a leech heart interneu- pyramidal neuron model. IEEE Trans Biomed Circ Syst 12:47–57
ron. Phys Rev E 84(1–8):041910 61. Selverston AI, Rabinovich MI, Abarbanel HDI, Elson R, Szucs A,
41. Marin B, Barnett WH, Doloc-Mihu A, Calabrese RL, Cymbalyuk Pinto RD et al (2000) Reliable circuits from irregular neurons: a
GS (2013) High prevalence of multistability of rest states and dynamical approach to understanding central pattern generators.
bursting in a database of a model neuron. PLoS Comput Biol J Physiol 94:357–74
9(1–12):E1002930 62. Shilnikov A, Cymbalyuk G (2005) Transition between tonic spik-
42. Masoller C (1994) Coexistence of attractors in a laser diode ing and bursting in a neuron model via the blue-sky catastrophe.
with optical feedback from a large external cavity. Phys Rev A Phys Rev Lett 94(1–4):048101
50:2569–78 63. Storace M, Linaro D, De Lange E (2008) The Hindmarsh–Rose
43. Megam NEB, Fotsin HB, Louodop FP, Kamdoum TV, Hilda CA neuron model: bifurcation analysis and piecewise-linear approxi-
(2016) Bifurcations and multistability in the extended Hindmarsh- mations. Chaos 18:033128
Rose neuronal oscillator. Chaos Solitons Fractals 85:151–63 64. Tuna M, Karthikeyan A, Rajagopal K, Alcin M, Koyuncu I (2019)
44. Mondal A, Kumar Upadhyay R, Ma J, Yadav BK, Sharma SK Hyperjerk multiscroll oscillators with megastability: analysis,
(2017) Bifurcation analysis and diverse firing activities of a modi- FPGA implementation and a novel ANNring-based true random
fied excitable neuron model. Cogn Neurodyn 13:393–407 number generator. AEU-Int J Electron Commun 112:152941
45. Morris C, Lecar H (1981) Voltage oscillations in the barnacle 65. Tapche RW, Njitacke ZT, Kengne J (2021) Complex dynamics of
giant muscle fiber. Biophys J 35:193–213 a novel 3D autonomous system without linear terms having line
46. Nayfeh AH, Balachandran B (1995) Applied nonlinear dynamics of equilibria: coexisting bifurcations and circuit design. FB Pelap
analytical, computational, and experimental methods. Germany Analog Integr Circ Signal Process 103(1):57–7
Wiley 66. Vaithianathan V, Veijun J (1999) Coexistence of four different
47. Njitacke ZT, Kengne J, Fotsin HB (2018) A plethora of behaviors attractors in a fundamental power system model. IEEE Trans Circ
in a memristor based Hopfield neural networks (HNNs). Int J Dyn Syst I 46:405–9
Control 7:36–52 67. Wouapi KM, Fotsin HB, Feudjio KF, Njitacke ZT (2019) Hopf
48. Njitacke Z, Kengne J (2018) Complex dynamics of a 4D Hopfield bifurcation, offset boosting and remerging Feigenbaum trees in
neural networks (HNNs) with a nonlinear synaptic weight: coex- an autonomous chaotic system with exponential nonlinearity. SN
istence of multiple attractors and remerging Feigenbaum trees. Appl Sci 1:1715
AEU-Int J Electron Commun 93:242–252 68. Wouapi KM, Fotsin H, Louodop FP, Feudjio KF, Njitacke TZ,
49. Njitacke ZT, Kengne J, Fonzin FT, Leutcha BP, Fotsin HB (2019) Djeudjo TH (2020) Various firing activities and finite-time syn-
Dynamical analysis of a novel 4-neurons based Hopfield neural chronization of an improved Hindmarsh-Rose neuron model under
network: emergences of antimonotonicity and coexistence of mul- electric field effect. Cogn Neurodyn. https://​doi.​org/​10.​1007/​
tiple stable states. Int J Dyn Control 7:823 s11571-​020-​09570-0
50. Njitacke ZT, Kengne J (2019) Nonlinear dynamics of three-neu- 69. Xu Y, Jia Y, Ge MY, Lu LL, Yang LJ, Zhan X (2018) Effects of
rons-based Hopfield neural networks (HNNs): remerging Feigen- ion channel blocks on electrical activity of stochastic Hodgkin-
baum trees, coexisting bifurcations and multiple attractors. J Circ Huxley neural network under electromagnetic induction. Neuro-
Syst Comput 28:1950121 computing 283:196–204
51. Njitacke ZT, Doubla IS, Mabekou S, Kengne J (2020) Hidden 70. Zhang S, Cui K, Zhang X, Shi X, Ge M, Zhao M, Xu G, Yan W
electrical activity of two neurons connected with an asymmetric (2018) Effect of transcranial ultrasonic-magnetic stimulation on
electric coupling subject to electromagnetic induction: coexist- two types of neural firing behaviors in modified Izhikevich model.
ence of patterns and its analog implementation. Chaos Solitons IEEE Trans Magn 54:5000204
Fractals 137:109785 71. Ying X, Yeye G, Guodong R, Jun M (2020) Dynamics and stochastic
52. Njitacke ZT, Doubla IS, Kengne J, Cheukem A (2020) Coexist- resonance in a thermosensitive neuron. Appl Math Comput 385:125427
ence of firing patterns and its control in two neurons coupled 72. Ying X, Minghua L, Zhigang Z, Jun M (2020) Dynamics and
through an asymmetric electrical synapse. Chaos 30:023101 coherence resonance in a thermosensitive neuron driven by pho-
53. Njitacke ZT, Sone ME, Fozin TF, Tsafack N, Leutcho GD, Tch- tocurrent. Chin Phys 29:098704
apga CT (2021) Control of multistability with selection of chaotic 73. Yong L, Wan-jiang X, Jun M, Faris A, Aatef H (2020) A new pho-
attractor: application to image encryption. Eur Phys J Spec Top tosensitive neuron model and its dynamics. Inf Technol Electron
230(7):1839–1854 Eng Front. https://​doi.​org/​10.​1631/​FITEE.​19006​06
54. Noble D (1960) Cardiac action and pacemaker potentials based
on the Hodgkin-Huxley equations. Nature 188:495 Publisher's Note Springer Nature remains neutral with regard to
55. Panahi S, Aram Z, Jafari S, Ma J, Sprott JC (2017) Modeling of jurisdictional claims in published maps and institutional affiliations.
epilepsy based on chaotic artificial neural network. Chaos Solitons
Fractals 105:150–6

13